.png)
Few things are as exhilarating as a well-executed AI demo. The technology is evolving so quickly that it feels like we're watching the future get built in real time.
And yet, the reputation of AI demos and POCs seems to be cratering. New products quickly capture the imagination before becoming punching bags. AI agents for code generation are the most polarizing thing on my Twitter feed. Enterprises are littered with failed AI POCs. And everybody still hates Google for lying to them about a duck. It's mean out there!
In an attempt to better understand the source of the underlying frustration, let's distinguish between two types of demos: the marketing demo and the progress demo.
Think of a great marketing demo like an immersive movie experience. When I saw Gravity in theaters I felt like I was in outer space for 2 hours. I stumbled out of the theater shell shocked and a little confused to be back on planet earth.
It takes a great deal of planning, design, and engineering to create a movie experience that allows the audience to experience outer space. This is how I think about marketing demos. Like Gravity, these demos are fundamentally pieces of theater that also require a great deal of planning, design, and yes, real engineering. Marketing demos spark the imagination. They can show us a glimpse of the future, even if that future isn’t ready to go into production.
Take Figure’s demo from March: the first time I watched it, my head exploded with the thought that we’re actually going to have robots in our kitchens doing our dishes! Do I care that this technology isn’t production ready right now? No! Seeing the future is great!
Problems arise when your expectations are in the wrong place. It's fine for a marketing demo to be full of smoke and mirrors. It's bad when the viewer thinks they're evaluating a real product. For that, we need an entirely different type of demo.
Marketing demos show you the future. Progress demos show you the present. There is no editing - just a transparent display of the progress along the critical path to production.
SpaceX launches are my favorite progress demos. Every big Starship launch seemingly brings us one step closer to Mars. Of course the rockets all explode and we haven’t gotten to Mars yet, but that’s to be expected! SpaceX isn’t claiming the rocket is production ready - they’re running test launches, gathering data, and showing their progress in real time.
The honesty of these demos makes it difficult to control the narrative - every inspiring launch by SpaceX produces more bad headlines about exploding rockets. But this honesty is also what makes progress demos so powerful. They’re an opportunity to level-set about capabilities and limitations. To get real world feedback about feasibility and development timelines.
So when you are tasked with building a POC, which of these demos are you being asked to build?
Here’s a conversation playing out over-and-over again:
Boss: Build a quick POC for [insert AI Project] to see if it’s even possible.
[couple of weeks later]
Build team: We did some prompt engineering using OpenAI’s API and actually got it to work pretty well. It’s not super reliable, but watch it get these examples right! This project is likely doable!!!
Boss: Awesome! How long until we can put this into production?
Build team: Uhhh…I have no idea. To build a production version we have to throw all of this out and start over with a more sophisticated approach.
Cue lots of frustration on both sides, followed by the project being abandoned.
Why? The question "is this even possible" is a trap. It feels like a smart way to get a quick answer, but in reality the project is being aimed at the scrapheap from the start. Companies have limited resources, and saying “this project is possible” is rarely enough to justify continued investment. Resources flow towards excitement or progress.
To set the stage for a successful POC the first step is to align on the goal. Generate excitement or show progress? Marketing demo or progress demo? More often than not you’ll want to aim for real progress. Here are some best practices for AI progress demos/POCs:
There’s a real place for marketing demos, but for POCs the safest bet is to stay on the critical path to production.
Fractional AI is hiring generalist software engineers to join our San Francisco based team. Get in touch with us if you want to help us take on the most challenging implementations of AI.
#AI #generativeAI #demo #hiring
Chris Taylor is the CEO of Fractional AI. Before launching Fractional AI, Chris was CEO of Xip, CRO of Wove, and held sales leadership positions at LiveRamp.
